This client is a global insurance company and a leader in specialty insurance writing business in both the Lloyd’s and the company markets. Describe Data was engaged to provide an independent assessment of the suitability of their current Marine Hull data for the purposes of data driven underwriting and pricing.
The aim was to test the hypothesis that their underwriting approach is effective and efficient, and that they have the appropriate quality and granularity of data required.
Describe Data was chosen to perform this work because we have extensive experience in the analysis of complex data sets and have delivered data analytics and technical solutions for insurance and reinsurance companies in the UK, Ireland and Bermuda.
The client used a number of data sources to price marine hull insurance, including:
These data sources were used to calculate exposure and also to estimate frequency and severity of claims.
Describe Data’s brief was to determine if the data sources were adequate to use for pricing. Furthermore we were also tasked with providing advice as whether the company could extract further value out of their currently available data to become a “best in class” underwriter and, if not, what course of action would make this possible.
We began by collating and validating the data sources that the client used for underwriting and pricing. This was not a trivial task.
We used a combination of the Lloyd’s List data that contains vessel-level technical information as well as a list of incidents categorised as Major or Minor severity. In addition, we used third-party data to add behavioural information such as regions of the world visited, the count of both total and unique ports visited, total distance travelled and so on.
To assess the ability of this data to improve risk models, we built a number of frequency models using an increasing number of variables. The baseline model contains just the year of account, along with the vessel type and vessel age.
We incrementally added a series of additional variables based on vessel behaviour, such as total port visits, unique port visits and total distance travelled.
To assess these models we used the outputs of the fitted models to predict the total incidents estimated by the model, using Monte Carlo simulation to get a distribution of predicted incidents. We compared this distribution of counts against the total count observed in the data.
Marine Pricing Model Outputs in R Markdown
